Last night I witnessed (and possibly participated in) a great example of my type of geek humour. It was a spontaneous collaborative joke, with 3 or 4 people making suggestions.
A modern couple gets married, and decides they will come up with an entirely new surname. They decide that this will be done during the ceremony, using the input of the guests present. This is done by ______.
To avoid extreme ridiculousness in their new name, they introduce a constraint, namely ______.
I can't remember the exact suggestions... but I'm curious to see what my LJ readers come up with.
I have long said that the differences between PT and BR Portuguese are very analogous to UK vs US English. Sometimes I reconsider this. For example, if you're a Wikipedian, there's some annoying differences between the Portuguese spoken in Portugal and Brazil. I now have slightly more sympathy for those who proposed to separate the BR and PT Wikipedias, while still strongly disagreeing with it.
Today I saw Carl Zimmer talk about the evolution of viruses, specifically how they cross the species barrier (between birds, pigs and humans). Although birds seem to be the (direct or indirect) source of all human viruses, and even though they harbor all the Hs from 1 to 16, it is pigs that seem to be good at promoting "viral sex", i.e. recombinations between different virus types. Apparently, industrial-scale pig-farming barns seem to be huge labs for viral creativity.
There is concern that the 60%-deadly H5N1 might evolve the ability to spread from human to human, but it still seems to be 13 mutations away from doing that. I found this number awfully precise, and upon questioning, he told me that these 13 genes are the ones that are always present when you look at the 'diff' between bird-virus and human-virus, i.e. right before and right after it migrates to humans. The "always", however, only refers to the great pandemics, of which there are 3... so I'm concerned about the statistical significance of this.
He showed us this awesome video from NPR (who knew they made videos!):
QOTD: "The virus lives in the tropics year-round, and comes up North during our winter. If it's any consolation, they come here to die"
I've just realized that using hash tables properly requires a notion of when two objects have equal value.
How come I never had to worry about this in the past?
---
I want my hash table to encode an equivalence relation (permutations of node labels), automatically dealing with my label-switching problem.
---
Do any of the normal languages/libraries let you pass an equality test, and adapt the hash function accordingly? I imagine you'd need to pass a function that brings the objects into a normal form.
---
I've installed R's hash library, but it seems to be pretty basic.
I'd love to see a plot showing "number of viruses exposed" vs "probability of infection", where subjects are injected with viruses in varying quantities.
Presumably a single virus (what do you call an individual virus?) isn't enough to make you sick most of the time. I imagine we'd be sick all the time if that were the case...
If you inject a single virus into bloodstream, is it most likely to get lost, get blocked or reproduce?
Can we develop immunity by low-grade exposure to a virus?
My question also applies to other microbial parasites.
Wow, this storm is so strong that I'm fearing for my electricity. TheWeatherNetwork says 35km/h but has a link to a Wind Warning:
WIND WARNING: Greater Vancouver Issued at 8:35 PM PST MONDAY 16 NOVEMBER 2009 SOUTHERLY WINDS 50 TO 70 KM/H WILL DEVELOP LATER THIS EVENING. THIS IS A WARNING THAT POTENTIALLY DAMAGING WINDS ARE EXPECTED OR OCCURRING IN THESE REGIONS.
Is there a cheap and easy way to harness wind power?
I'm not expecting the trees to have many leaves left in the morning!
According to the sources I've read, Gibbs sampling doesn't really tell you how to sample... it's just divide-and-conquer: you still need to sample each variable from its full conditional.
In my problem, that would be more trouble than it's worth, since no denominators are canceling. So I'm doing good old Metropolis-Hastings, with a proposal that makes its trajectory resemble Gibbs: it randomly chooses one variable to modify, and then randomly chooses a value for that variable. In other words, the proposal is uniform over the neighbors in the "generalized hypercube"*.
I can easily compute the posterior. In traditional MCMC, I think you would weight the samples by how often they appear. But doesn't it make more sense to directly compute the posterior in all sampled models?
Should I be using MCMC at all? How else am I going to find the set of high-probability models? (Maybe what I want is a mode-oriented stochastic search, as my EA project did.
Believe it or not, it's my first time actually implementing MCMC.
Also, I'd like to know what proportion of the posterior mass my sampled models account for... but this is probably VeryHard to do if I can't enumerate the models (otherwise we could know whether the chain has mixed).
* - what do you call a non-binary hypercube? I mean: let each node be a string in {0,...,k-1}^n, and neighborhood relation is defined by set of strings that differ in exactly 1 character. When k=2, we get the n-dim hypercube. What's the word when n>2?.
Having a TV this week is putting me more in touch with ordinary folks, for whose sake science maxims are subverted, e.g. "The NuWave Oven uses all 3 forms of heat transmission: conduction, convection and infrared!!". Everyone knows that "radiation" is bad for you...
As you may know, MRI used to be called NMRI, but medicine and "nuclear" don't go together.
After my place is painted, I will try to look for a place to a print a big poster of the 500-year-old painting, the Biblically-inspired Garden of Earthly Delights.
If it had been painted by Nostradamus, I would say he was envisioning Burning Man.
Difference between sun time and clock time during daylight saving time:
I always found strange how Spain insists on being GMT+1 (which is perfect for more Eastern countries such as Austria and Česko) despite being West of Britain... Presumably, the UK, being powerful and isolated by water, didn't feel the pressure to conform. I wonder when and how Portugal decided to join the UK with GMT.
---
Here's another, which I haven't fully grasped yet:
Equation of time << The equation of time is the difference between apparent solar time and mean solar time, both taken at a given place (or at another place with the same geographical longitude) at the same real instant of time. >>
The equation of time — above the axis the dial will appear fast, and below the dial will appear slow.
Does having a virus make you more or less susceptible to other viruses? I can see it going both ways... If your immune system is activated, it might protect you more in general (at the cost of allergy-like symptoms, such as mucus)... or it might protect you specifically against what you have, at the expense of other battles.
Of course, there's viruses that destroy your immune system, such as HIV, but that's not what I'm talking about.
\includegraphics[height=60mm]{study2 - r varies/rtrue=10,r=10,sMod=[3,4,5,6],rs=[10,5,2.5],score=loglik.pdf} breaks because the first "." makes the "5" be interpreted as a file extension.
\includegraphics[height=60mm]{study2 - r varies/rtrue=10,r=10,sMod=[3,4,5,6],rs=[10,5,2;5],score=loglik.pdf} compiles just fine and displays the figure, but it dumps the filename in my document, even if I do it inside a 'figure' environment.
Since computing my confidence interval is too time-consuming, I'd like to use some sort of errorbar that means "insert errorbar here". The point is to illustrate that what I'm plotting is not an exact function... but maybe that's obvious and my pedanticism is going too far here.
Sometimes I make two-color scatterplots with o-shaped points. Occasionally, when I have a few thousand points, they bunch up and it really matters which color was plotted first: I can't see behind the color that was plotted last... except when Acrobat refreshes its window. ...which of course means it's possible to manipulate the elements of a PDF. But how?
Ideally, these points would be plotted at a random order (independent of color) AND would be semi-transparent. I'm not sure how to accomplish this easily with R's 'plot'.
Idea for Firefox: before opening a risky URL, give a crash warning: "Are you sure?". You could learn from experience which are safe URLs or plugin types are safe... you could even explain to the user why it thinks it's a crash risk.
Another idea for Firefox: make each tab its own sandbox!
Windows: The program stops responding. Step 1: Try to exit by clicking "x" and/or File menu. This fails, pointer turns into an hourglass. Step 2: Ctrl-Alt-Del and "End Now". This fails. Step 3: Wait, try again, wait, try again, like 30 times. This fails. Step 4: Right-click, select "End Process Tree" and accept the dire consequences. This fails. Step 5: Reboot by going to "Restart". Since Windows waits for processes to shut down, this too fails. ... until magically I notice that it's restarting.
It feels like Windows gradually forces programs to shut down, by stronger measures. I wonder what's going on.
Why does "End now" not mean anything anymore... IMHO, we should go no further than Step 2.
